lokalise-core-workflow-b
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides instructions and code snippets that use
execSyncto run shell commands. Specifically, it demonstrates usingcurlto download translation bundles andunzipto extract them into the project directory. It also includes a Vite plugin example that executesnpm run i18n:pullduring the build process. - [EXTERNAL_DOWNLOADS]: The skill facilitates the download of translation assets from Lokalise's infrastructure. It retrieves authenticated S3 bundle URLs via the Lokalise API and downloads them using
curloraxiosfor integration into the application. - [INDIRECT_PROMPT_INJECTION]: The skill processes external translation data, which establishes a surface for indirect prompt injection. 1. Ingestion points: Translation files (JSON, XLIFF, PO) fetched from the Lokalise API as shown in
SKILL.mdandreferences/implementation-guide.md. 2. Boundary markers: None present; the examples do not include delimiters or warnings to ignore instructions embedded within translation strings. 3. Capability inventory: The skill usesexecSyncfor subprocess execution,fs.writeFileSyncandAdmZipfor file system writes, andaxios/curlfor network requests. 4. Sanitization: No sanitization or validation of the translation content is performed before writing files or generating TypeScript types.
Audit Metadata